Application of neural networks to identify trolls in social networks.

In this paper we developed and tested a new algorithm of detecting in social networks users (so-called trolls) who behave in an insulting and provocative way towards other users. In order to detect trolls it is proposed to unite users in groups where all the members have a similar way of communicating. Defining the number of group and distributing the users into these groups is carried out automatically due to application of neural networks of special type - Kohonens self-organized maps. As for users characteristics according to which the distribution into groups might be done we suggest using such data as the number of comments, the average comment length and indicators determining the emotional state of the user (the frequency of encountering certain characters in comments).